Topics

Can AI truly learn and reason? DeepProbLog bridges neural networks and logic, combining perception, uncertainty, and symbolic reasoning to create smarter, explainable AI.

LLMs hallucinate confidently. Uncertainty estimation flags errors in NLP & code—strong on open-source models, misses nuances in GPT-4o. A plug-and-play step toward safer, self-aware AI.

Multi-Objective Causal Bayesian Optimization revolutionizes decision-making in healthcare and engineering, integrating causal graphs with Bayesian methods to efficiently navigate trade-offs, ensuring smarter, interpretable solutions for complex systems.

Tired of AGI leaderboard illusions? A data scientist reveals the real test: can it invent Bitcoin in 2007, teach quantum physics in Swahili, and know when to ask for help?

GOLLuM trains LLMs using Bayesian uncertainty signals, raising high-yield Buchwald-Hartwig discovery from 24% to 43% in 50 experiments across 19 benchmarks.

MIT researchers developed eta-learning, which enforces statistical constraints from qualitative knowledge during training to generate plausible extremes even when none appear in the data.

An information-theoretic framework quantifying intensional inheritance, enabling AI systems to model conceptual meaning, probabilistic relationships, and human-like reasoning beyond simple example-based hierarchies.